بتل کد Battle Codeدارد باز می‌شود…

جواب «Step 4: Learn from One Game» چالش 4 از جلسه‌ی 61

سؤال

قدم ۴: در play_episode هر حرکت عامل را به شکل (sticks, t) به history اضافه کن، و بدنه‌ی learn(history, reward, lr) را بنویس: هر Q[key] تاریخچه را کمی به سمت reward ببرد. برنامه یک بازی انجام می‌دهد (عامل می‌بازد) و حدس‌های بعد از یادگیری را چاپ می‌کند. خروجی، بعد از خط‌های قدم‌های قبل، باید این باشد:
reward: -1 | agent moves: [(10, 1), (8, 1), (5, 1), (3, 1)]
after learning: [-0.5, -0.5, -0.5, -0.5]

راهنما

در شاخه‌ی عامل: history.append((sticks, t)). در learn: for key in history: Q[key] += lr * (reward - Q[key]).

یک جواب درست

import random

START = 10


def moves(sticks):
    return [t for t in (1, 2, 3) if t <= sticks]


def play_game(first, second):
    sticks = START
    players = [first, second]
    turn = 0
    while True:
        t = players[turn](sticks)
        sticks = sticks - t
        if sticks <= 0:
            return turn
        turn = 1 - turn


def take_one(sticks):
    return 1


def take_most(sticks):
    return max(moves(sticks))


print(moves(10), moves(2), moves(1))
print("winner:", play_game(take_one, take_most))


def random_player(sticks):
    return random.choice(moves(sticks))


random.seed(1)
wins = [0, 0]
for g in range(1000):
    wins[play_game(random_player, random_player)] += 1
print("random vs random:", wins)


Q = {}
for s in range(1, START + 1):
    for t in moves(s):
        Q[(s, t)] = 0.0


def choose(sticks, epsilon):
    if random.random() < epsilon:
        return random.choice(moves(sticks))
    best = moves(sticks)[0]
    for t in moves(sticks):
        if Q[(sticks, t)] > Q[(sticks, best)]:
            best = t
    return best


print("Q-table size:", len(Q))
Q[(7, 3)] = 0.8
print("greedy at 7:", choose(7, 0.0), "| greedy at 5:", choose(5, 0.0))
Q[(7, 3)] = 0.0


def play_episode(epsilon, agent_first):
    sticks = START
    history = []
    agent_turn = agent_first
    while True:
        if agent_turn:
            t = choose(sticks, epsilon)
            history.append((sticks, t))
        else:
            t = random_player(sticks)
        sticks = sticks - t
        if sticks == 0:
            reward = 1 if agent_turn else -1
            return reward, history
        agent_turn = not agent_turn


def learn(history, reward, lr):
    for key in history:
        Q[key] += lr * (reward - Q[key])


random.seed(4)
reward, history = play_episode(0.2, True)
print("reward:", reward, "| agent moves:", history)
learn(history, reward, 0.5)
print("after learning:", [Q[key] for key in history])

اول خودت امتحان کن؛ جواب‌های دیگر هم قبول است. برگرد و خودت حلش کن